Spaces:
Running
Running
| # https://huggingface.co/spaces/amendolajine/OPIT | |
| # Here are the imports | |
| import logging | |
| import gradio as gr | |
| import fitz # PyMuPDF | |
| from transformers import BartTokenizer, BartForConditionalGeneration, pipeline | |
| import scipy.io.wavfile | |
| import numpy as np | |
| # Here is the code | |
| # Initialize logging | |
| logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s') | |
| # Initialize tokenizers and models | |
| tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn') | |
| model = BartForConditionalGeneration.from_pretrained('facebook/bart-large-cnn') | |
| synthesiser = pipeline("text-to-speech", "suno/bark") | |
| def extract_abstract(pdf_bytes): | |
| try: | |
| doc = fitz.open(stream=pdf_bytes, filetype="pdf") | |
| first_page = doc[0].get_text() | |
| start_idx = first_page.lower().find("abstract") | |
| end_idx = first_page.lower().find("introduction") | |
| if start_idx != -1 and end_idx != -1: | |
| return first_page[start_idx:end_idx].strip() | |
| else: | |
| return "Abstract not found or 'Introduction' not found in the first page." | |
| except Exception as e: | |
| logging.error(f"Error extracting abstract: {e}") | |
| return "Error in abstract extraction" | |
| def process_text(uploaded_file): | |
| logging.debug(f"Uploaded file type: {type(uploaded_file)}") | |
| logging.debug(f"Uploaded file content: {uploaded_file}") | |
| try: | |
| with open(uploaded_file, "rb") as file: | |
| pdf_bytes = file.read() | |
| except Exception as e: | |
| logging.error(f"Error reading file from path: {e}") | |
| return "Error reading PDF file", None | |
| try: | |
| abstract_text = extract_abstract(pdf_bytes) | |
| logging.info(f"Extracted abstract: {abstract_text[:200]}...") | |
| except Exception as e: | |
| logging.error(f"Error in abstract extraction: {e}") | |
| return "Error in processing PDF", None | |
| try: | |
| inputs = tokenizer([abstract_text], max_length=1024, return_tensors='pt', truncation=True, padding="max_length") | |
| summary_ids = model.generate( | |
| input_ids=inputs['input_ids'], | |
| attention_mask=inputs['attention_mask'], | |
| pad_token_id=model.config.pad_token_id, | |
| num_beams=4, | |
| max_length=45, | |
| min_length=10, | |
| length_penalty=2.0, | |
| early_stopping=True, | |
| no_repeat_ngram_size=2 | |
| ) | |
| summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True) | |
| words = summary.split() | |
| cleaned_summary = [] | |
| for i, word in enumerate(words): | |
| if '-' in word and i < len(words) - 1: | |
| word = word.replace('-', '') + words[i + 1] | |
| words[i + 1] = "" | |
| if '.' in word and i != len(words) - 1: | |
| word = word.replace('.', '') | |
| cleaned_summary.append(word + ' and') | |
| else: | |
| cleaned_summary.append(word) | |
| final_summary = ' '.join(cleaned_summary) | |
| final_summary = final_summary[0].upper() + final_summary[1:] | |
| final_summary = ' '.join(w[0].lower() + w[1:] if w.lower() != 'and' else w for w in final_summary.split()) | |
| speech = synthesiser(final_summary, forward_params={"do_sample": True}) | |
| audio_data = speech["audio"].squeeze() | |
| normalized_audio_data = np.int16(audio_data / np.max(np.abs(audio_data)) * 32767) | |
| output_file = "temp_output.wav" | |
| scipy.io.wavfile.write(output_file, rate=speech["sampling_rate"], data=normalized_audio_data) | |
| return final_summary, output_file | |
| except Exception as e: | |
| logging.error(f"Error in summary generation or TTS conversion: {e}") | |
| return "Error in summary or speech generation", None | |
| iface = gr.Interface( | |
| fn=process_text, | |
| inputs=gr.components.File(label="Upload a research PDF containing an abstract"), | |
| outputs=["text", "audio"], | |
| title="Summarize an abstract and vocalize it", | |
| description="Upload a research paper in PDF format to extract, summarize its abstract, and convert the summarization to speech. If the upload doesn't work on the first try, refresh the page (CTRL+F5) and try again." | |
| ) | |
| if __name__ == "__main__": | |
| iface.launch() |